In this paper, a hybrid multi-dimensional image segmentation algorithm is proposed, which combines edge and region-based techniques through the morphological algorithm of simplification, connected operators and watersheds. At first, in order to compute an accurate estimate of the image gradient, an edge- preserving connected operator is used as a preprocessing stage to simply the image and reduces noise. Then, applying the watershed transform on image gradient magnitude produces an initial partitioning of the image into primitive regions. This initial segmentation is the input to a bottom-up region merging process that produces the final segmentation. The latter process is iterative and uses the region adjacency graph (RAG) representation of the image region and the relationship between regions. At each merging step, the most similar pair of regions is determined, the regions are merged and the RAG is updated. At the steps of image simplification, watershed transform and region merging, a hierarchical queue is used and information is propagated only in the relevant image parts. As a result, the efficiency of the algorithm turns out to be very high. The final segmentation provides one-pixel wide, closed, and accurately localized contours. Thus, we can get small or long-and-thin parts of the image, nor can the traditional watershed-plus-markers method. Furthermore, a general region-merging algorithm is present in this paper. Based on different merging order, merging criterion and region model, many operators can be created. Benefiting from the filtering and segmentation viewpoints, it can be used as both segmentation and filter tool. The new connected operators are not only self- dual, but also can reduce transition regions. If more complex function is used as merging order, new connected operators can be created which have no assumption that objects composing scene are either bright or dark image components. Experimental results and analysis obtained with 2D image are presented in the paper.
修改评论